FAQs

POD is Penguin Computing's high-performance computing on demand service in the cloud. POD provides a secure infrastructure plus the hands-on services and HPC expertise to enable you to run applications at performance levels at or exceeding that of an on-premise HPC data center. POD scales with your need and gives you a resource where you only pay for what you use.

POD is targeted at analysts, researchers, scientists, students and engineers who need high-performance computing capabilities without the capital expense and effort required to acquire and operate HPC clusters or who just require surge capacity for time-critical analyses. POD is intended both for end-users as well as for ISV's and service-providers.

There are several differences. First, POD is geared towards applications that thrive in an HPC environment which are starved for performance on a virtualized cloud computing environment. Parallel workloads common in the HPC space can be extremely sensitive to latency and I/O performance, particularly as they scale. Secondly, POD is designed with both direct-attached scratch storage for HPC applications and high-speed, parallel file storage for mission critical data. Fast access to storage is critical for HPC workloads to run efficiently. Third, POD is designed so that user's can scale up the cores and memory they need, only limited by the physical cores available on the cluster.

POD uses Intel and AMD HPC clusters located in a variety of cities in the US. These clusters provide POD users with their own personal, persistent and secure virtualized login-node plus a dynamically scalable compute cluster where the user's applications run directly on the compute node hardware (non-virtualized). Both GigE and Infiniband compute network fabrics are available.

Because of the non-virtualized and highly scalable nature of the POD compute environment, the wall clock time for job execution (time-to-result or TTR) is significantly reduced compared with traditional cloud-computing offerings. We have seen up to a 30x improvement in TTR.

The POD infrastructure includes both AMD and Intel HPC clusters with the choice of a GigE or Infiniband network, direct-attached high-speed storage, persistent file storage for user data, and a high-speed parallel file storage system for use with IO intensive applications. User's HPC applications run directly on the compute nodes without virtualization so performance is maximized.

Each POD has a soft limit of 480 cores available to a customer. However, POD can be rapidly scaled by Penguin so POD can grow its capacity as needed.

A number of servers in the POD infrastructure offer GPU capabilities.

POD can be rapidly configured or scaled by adding specialized compute environments (such as high-memory capacity nodes) to meet specific customer requirements. It is important to keep in mind that, as a system integrator, Penguin has been providing HPC solutions including servers, storage, software environment and professional services for many years. Depending on the type of hardware required, a minimum use commitment may be required.

POD is easy to access, use and submit jobs. You simply submit your Torque or GridEngine script from your login node.

Our facility has the bandwidth needed to allow the transfer of reasonably large data files directly to or from POD over the Internet. POD also includes a "disk caddy" service that allows the transfer of files overnight.

It is very easy to get started with POD - see Getting Started.

Not with the general offering, though SLAs and QoS can be implemented for environments that require it. We do guarantee that if a job fails because of a POD hardware failure, then that job can be rerun at no cost. POD is designed to run at 50% average utilization per month, maintaining sufficient headroom to ensure that queue wait times are minimized.

The POD software environment is continually updated to remain current and provide the latest tools to our customers.

The POD infrastructure is CentOS based which is 100% compatible to RedHat. POD runs on X86-64 hardware. Complete GNU and Intel compilers and libraries are available.

Customers can bring their code to build on POD. Penguin has been aiding customers in moving their applications on to Penguin clusters for years and has a tremendous amount of experience here.

Penguin knows that security and privacy are one of your top concerns when it comes to using a HPCaaS or compute on demand vendor. We believe in strong security policies and best practices to protect your data, provide network access only to authorized users, and identity security to ensure any access is properly validated. Our policies and best practices are followed to ensure security of your data while it's on our system.

Penguin's knowledge of high performance computing and Linux, coupled with our best practices of security audits and system scans, provides the best protection of your data offered by the industry today.

Yes. POD offers several SaaS applications, including RealityServer from mental images, LS-Dyna from LSTC plus BioScope and LifeScope from Life Technologies. Others can be accessed from POD as well such as MatLab, STAR-CD and STAR-CCM+. Please contact us if you need access to a particular commercial application.

POD charges are on a price per core and based on actual hours used. Core hour fees are calculated on a 1/1000 of an hour accuracy.

POD is priced to provide a significant improvement in price/performance for HPC applications when compared to running on traditional cloud computing offerings. Keep in mind that for HPC applications, time to results (TTR) or wall clock time is a key metric in measuring value. POD's prices per core hour are similar to many cloud computing offerings, but it is common for the price per job to be less since POD executes jobs much faster. Users pay only for the core hours that they use.

Cost to run a job on POD can vary depending on resources available and model size. Penguin Computing has expertise in high performance computing and applications so we can provide an estimate, but cannot provide a definitive answer on the cost to run your job.